How this is built

Method

This register exists to answer questions about retirement and welfare plans without inventing anything. That constraint shapes every part of the pipeline.

Source and refresh

Filings come from the U.S. Department of Labor EBSA Form 5500 public disclosure datasets. Ingestion runs monthly, because DOL refreshes the 2009-and-later files around the first of each month while older years are static. Each run resolves the currently published dataset, skips work when that exact version was already ingested, streams the file rather than buffering it, and records what it saw. The refresh date shown next to the data is the dataset's, not the day you happened to load the page.

Columns change; the register does not

Form 5500 column names differ between filing years. Nothing in this system hard-codes a column name. Every field is resolved through an alias map that lists the known spellings per year, and any column that is not mapped is preserved verbatim alongside the row, so no filed information is discarded. When a year introduces a column the map has never seen, the run records a warning instead of silently dropping the field.

Missing is not zero

If a plan did not file an item, this site says so in words and never prints a zero. Absence usually means the item is not collected on that form at all — Form 5500-SF, for instance, carries no Schedule C service-provider compensation detail. Treating that gap as a fee of zero would make the smallest filers look the cheapest, which is exactly the wrong conclusion.

Fees only as a position in a cohort

Compensation figures are never presented as a standalone verdict. They appear as a percentile inside a cohort you can state out loud — filing year, plan type, size band, state — and always with two counts: how many filings are in the cohort, and how many of them actually filed the measure being compared. Percentiles over a filed subset are only meaningful with that coverage visible.

Ratios such as compensation per participant, or compensation as a share of assets, are plain arithmetic over filed inputs. They are labelled as derived, and every derived result carries a logic version so the same question asked later returns a comparable answer.

What this is not

No value here is described as reasonable, high, low or excessive. A filing is not an endorsement of a plan or a provider, and nothing on this site is investment, tax, legal or fiduciary advice. Interpretation belongs to the reader, with the cohort and coverage in plain sight.